MCP Server for Cursor: Connect ToolYour Tools

Cursor agents can call remote tools over the Model Context Protocol (MCP). ToolYour provides a remote MCP server at https://api.toolyour.com/mcp so those agents can run API-backed utilities—document conversion, SEO checks, text operations, and more—with the same API key and monthly credits as the REST API.

This guide covers Cursor setup, the recommended agent loop (plan_task → solve_task → verify_task), what is (and is not) on MCP, shared billing with REST, and common connection failures.

What ToolYour MCP adds to Cursor

Without a remote MCP server, agents either invent fragile HTTP wrappers or skip real utility work. ToolYour MCP gives Cursor:

• One production endpoint: https://api.toolyour.com/mcp
• Auth with your existing ty_ key (X-Api-Key)
• Meta-tools for planning, solving, playbooks, verification, and discovery
• Access to API-backed tools only (not every browser-only page)

You keep writing goals in natural language inside Cursor. The server routes to workflows or tools, returns structured job reports, and bills successful runs from the same credit pool your REST scripts already use.

Full product context and discovery links live on the ToolYour MCP page.

Prerequisites (account and API key)

1. Create a developer account on ToolYour.
2. Open the API keys dashboard and create a key (ty_...).
3. Keep the key in a secret store—treat it like a production credential.

The same key works for REST and MCP. You do not need a second MCP-only key or a separate MCP subscription. If you already called REST from a backend, reuse that key (or create a labeled key for the agent machine).

Cursor mcp.json configuration

Add ToolYour to Cursor’s MCP config (exact file location depends on your Cursor version—user or project mcp.json):

{
  "mcpServers": {
    "toolyour": {
      "url": "https://api.toolyour.com/mcp",
      "headers": {
        "X-Api-Key": "ty_YOUR_KEY"
      }
    }
  }
}

Restart or reload MCP servers in Cursor so the agent can see ToolYour tools. Confirm the URL is exactly https://api.toolyour.com/mcp (the api.toolyour.com host)—do not invent an alternate MCP hostname. Streamable HTTP variants are documented on the MCP setup page when your client requires them; the default URL above is the canonical remote endpoint for most Cursor setups.

First agent workflow with plan_task and solve_task

Recommended loop:

1. plan_task (free) — pass a plain-language goal; get a ranked plan and a credit estimate without executing.

2. solve_task — execute the goal (default compact job report). Fuzzy matching and confidence gating apply; ambiguous goals return status: "suggest" with ranked suggestions instead of a wrong run (suggestions without execution are free).

3. verify_task — after you change a site or file, re-check against a prior baseline for deltas only.

Example goal shape (conceptual—your client wraps the MCP tool call):

{
  "goal": "Convert report.docx to PDF",
  "input": {
    "url": "https://yoursite.com/report.docx"
  },
  "responseMode": "compact"
}

For curated multi-step skills, use run_playbook after planning. When you already know a specific tool, use discover_tools → get_tool_schema → invoke_tool. Long-running work may expose a run id—poll with get_run when the docs say a job is async.

Do not treat solve_task as fully autonomous for every goal. When you see suggest, read the hint and example goals, clarify the request or supply missing input, then call again.

What tools are available (API-backed only)

MCP exposes tools that are API-backed in the product catalog—file conversion, documents, SEO/site utilities, text utilities, calculators, and AI utilities that ship with an API. Website-only browser pages are not exposed over MCP.

The public catalog is large (200+ tools across the site). That number is not the MCP tool count. Say “API-backed tools via MCP,” not “200+ tools on MCP.”

Agents can browse with discovery meta-tools without spending credits. Execution bills. Prefer specific goals (“compress this PNG”, “audit Core Web Vitals for this URL”) over vague “run all SEO tools” requests.

Quota and pricing (shared credits with REST)

Billing is credit-based and shared with REST:

• Free: 500 credits per month; tools cost 1–10 credits per successful run.
• Throughput on Free: 5 successful tool runs per rolling 60-second window.
• Free on MCP: catalog browse, plan_task, and suggestions without execution.
• Billed: successful tool runs through solve_task, invoke_tool, playbook/workflow steps, and similar execution paths.

REST calls and MCP runs draw from the same monthly credits. MCP is not billed as an add-on subscription. When you hit the cap, further billable runs are blocked until you upgrade or the monthly reset—no pay-as-you-go overage. See pricing.

Heavy SEO jobs and document pipelines spend more credits per call than light text utilities. Use plan_task estimates to budget before you execute in a loop.

Troubleshooting connection issues

| Symptom | What to check | |--------|----------------| | Tools missing in Cursor | Config saved, MCP reloaded, URL is https://api.toolyour.com/mcp | | Auth errors | Key starts with ty_, header is X-Api-Key, key not revoked | | suggest instead of a run | Goal too vague—use plan_task output / suggestions to refine | | Unexpected credit use | Confirm you executed a tool (not only browse/plan_task) | | Rate limited on Free | Wait for the rolling minute window or upgrade |

Still stuck? Re-read the MCP setup page and confirm the key works with a simple REST call from the developers hub. Check the dashboard usage graph after a test invoke so you know the key is attached to the account you expect.

FAQ

Q: What is the MCP endpoint for Cursor?
A: https://api.toolyour.com/mcp with X-Api-Key: ty_YOUR_KEY.

Q: Should agents call solve_task first?
A: Prefer plan_task (free) first, then solve_task. Use verify_task after changes. Ambiguous goals may return status: suggest.

Q: Is MCP a paid add-on?
A: No. MCP shares the REST monthly credit quota. Free includes 500 credits/month.

Q: Does browsing the catalog use credits?
A: No. Catalog exploration and plan_task are free. Running tools consumes 1–10 credits each.

Q: Are all ToolYour browser tools on MCP?
A: No. Only API-backed tools are exposed.

Q: Where do I create the API key?
A: In the API keys dashboard.

Q: Where is pricing?
A: https://www.toolyour.com/pricing.

Q: Can I use the same key in CI and Cursor?
A: Yes, but prefer separate keys labeled by environment so you can revoke an agent key without breaking backends. Credits still share one monthly pool per account.

Next steps

Connect the config above, run plan_task on a concrete goal, then solve_task. Keep developers/mcp bookmarked for discovery URLs and deeper agent patterns. When you outgrow Free, upgrade on pricing without changing your MCP URL or auth header shape.